Google Launches Gemini 2.5 Pro Preview: The New Standard for AI Development | Generative ai financial services | Microsoft generative ai tools free | Generative ai benefits for business examples | Turtles AI

Google Launches Gemini 2.5 Pro Preview: The New Standard for AI Development

Updated Gemini model improves programming, web aesthetics, and video understanding, previewing announcements expected at I/O 2025

Isabella V7 May 2025

Google Launches Gemini 2.5 Pro Preview: The New Standard for AI Development

Google has introduced Gemini 2.5 Pro Preview (I/O Edition), an advanced version of its AI model, available on the Gemini API, Vertex AI, AI Studio, and the Gemini chatbot app. The update, which comes ahead of Google’s annual I/O conference, introduces significant improvements in programming capabilities and multimodal understanding, placing it at the top of benchmarks such as WebDev Arena and VideoMME.

Key Points:

High performance: Gemini 2.5 Pro Preview outperforms WebDev Arena and VideoMME, demonstrating advanced capabilities in building web apps and understanding video.
Developer improvements: The model offers fewer errors in function calls and an improvement in the rate at which they are triggered, addressing developer feedback.
Accessibility: Available on Gemini API, Vertex AI, AI Studio, and the Gemini chatbot app for web and mobile, priced at Gemini 2.5 Pro.
Industry Competition: The launch comes ahead of Google I/O, where the company will showcase additional AI models and tools amid intense competition from rivals like OpenAI and xAI.

Google has anticipated the launch of its updated AI model, Gemini 2.5 Pro Preview (I/O Edition), bringing significant improvements in programming and multimodal understanding. Available through Gemini API, Vertex AI, AI Studio, and the Gemini chatbot app, the version stands out for its improved performance in benchmarks like WebDev Arena and VideoMME.

Gemini 2.5 Pro Preview (I/O Edition) represents a significant step in the evolution of Google’s AI models, providing enhanced tools for developers and users, and cementing the company’s position in the competitive AI landscape.

Google Launches Gemini 2.5 Pro Preview: The New Standard for AI Development | Generative ai financial services | Microsoft generative ai tools free | Generative ai benefits for business examples | Turtles AI

Goldman Sachs hires AI engineer Devin to assist its developers

The bank is testing an autonomous software agent capable of handling repetitive technical tasks, aiming to increase efficiency and introduce a hybrid workforce supervised by humans

Black Panther II, the robot dog that defies the clock: 100 meters in 13.17 seconds

The quadruped developed by Chinese startup Mirror Me surpasses previous records thanks to lightweight materials, artificial intelligence and a structure inspired by animal biomechanics

OpenAI still delays open-source model as pressure from competitors grows

OpenAI’s first open-weight model launch postponed indefinitely due to security concerns, while Moonshot AI unveils Kimi K2, outperforming GPT-4.1 in multiple key benchmarks

Moonshot AI’s Kimi K2 outperforms GPT-4 in testing: Open-source model with 1 trillion parameters

High performance in programming and reasoning benchmarks, a Mixture-of-Experts architecture, and stable training with MuonClip. Available in both basic and optimized versions, with a low-cost API for wider adoption

A new public and multilingual language model is born in Switzerland

ETH Zurich, EPFL and CSCS announce an open-source LLM trained on public infrastructure, capable of understanding over 1,000 languages and available under an open license for research, industry and institutions

Huawei Rejects Allegations of Plagiarism on AI Pangu Pro Model

Company denies similarities to Alibaba’s Qwen 2.5 after HonestAGI analysis, claims independent development and adherence to open source licenses

Hugging Face Launches Reachy Mini, the Open Source Desktop Robot for Developers

Available in Lite and Wireless versions, the robot is programmable in Python, integrated with Hugging Face Hub and designed for custom AI projects

Moonshot AI’s Kimi K2 outperforms GPT-4 in testing: Open-source model with 1 trillion parameters

High performance in programming and reasoning benchmarks, a Mixture-of-Experts architecture, and stable training with MuonClip. Available in both basic and optimized versions, with a low-cost API for wider adoption

A new technique for deceiving chatbots has been discovered: security filters under information stress

An automated attack confuses AI models with complex prompts, bypassing ChatGPT and Gemini’s protection systems and raising questions about their reliability in scenarios

Grok arrives on Tesla: xAI’s AI ready to debut in vehicles with Hardware 3

Elon Musk announces the upcoming integration of the Grok chatbot into Tesla cars, with customizable voice modes and advanced features, as questions arise about content moderation and restricted access to the latest models

See also

OpenAI’s Bug Bounty Program

Presti: AI transforming furniture product photos

Presti, a French startup, is transforming the furniture sector using AI, enabling the creation of realistic photos from single product images, saving brands time and costs.

Samsung and LG invest in edtech with inclusion of Ai digital textbooks in classrooms